skills/simota/agent-skills/vision/Gen Agent Trust Hub

vision

Pass

Audited by Gen Agent Trust Hub on Oct 22, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill functions as a hub for design direction, ingesting context, user research, and persona data from upstream agents like 'Field', 'Compete', 'Spark', 'Echo', and 'Vitrine'. This multi-source ingestion creates an indirect prompt injection surface. While the skill outlines grounding checks to ensure brand and persona alignment, it lacks explicit sanitization or strict boundary delimiters to prevent malicious instructions embedded in these external data inputs from influencing the agent's behavior.
  • [COMMAND_EXECUTION]: The 'multi' recipe implements a tri-engine architecture that dispatches tasks to external CLI binaries, specifically codex, agy (Antigravity), and claude. This capability involves shell command execution through a shared CLI adapter described in internal common files. These tools are invoked to generate parallel design directions, but they represent a command execution footprint on the local system.
  • [EXTERNAL_DOWNLOADS]: The skill's documentation contains numerous references to external technology and design resources. These include official Apple domains for human interface guidelines and news, as well as well-known technology journalism sites like MacRumors, 9to5Mac, MacStories, and Six Colors. These are used neutrally for trend verification and case study analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 22, 2026, 01:06 PM